| |
| |
| """The oracle for bankml's Rust mindXtrain (bankML/train/): mindXtrain's own Python, run on every persona in mindX |
| and cryptoAGI plus edge cases (author stage: data/scripts.py), and on 3,000 randomized before/after/baseline sets |
| (score stage: eval/imprint.py, lexical path). Run it with mindXtrain's interpreter: |
| ~/mindxtrain/.venv/bin/python testing/train_oracle.py [~/mindxtrain] |
| Writes .models/oracle-train/script.jsonl and imprint.jsonl.""" |
| import json, random, sys, tempfile |
| from pathlib import Path |
|
|
| MXT = Path(sys.argv[1] if len(sys.argv) > 1 else Path.home() / "mindxtrain").expanduser() |
| sys.path.insert(0, str(MXT)) |
| from mindxtrain.data.scripts import Exchange, build_script_rows, derive_training_params, persona_from_dict, write_script_jsonl |
| from mindxtrain.eval import imprint |
|
|
| out = Path(__file__).resolve().parents[1] / ".models" / "oracle-train" |
| out.mkdir(parents=True, exist_ok=True) |
| home = Path.home() |
|
|
| |
| files = sorted((home / "mindX/mindx/godel/mindxtrain/personas").glob("*.json")) + sorted((home / "mindX/mindx/godel/mindxtrain/personas").glob("*.persona")) |
| for d in ("savante", "jaimla", "luvai", "codephreak", "simplecoder"): |
| files += sorted((home / "cryptoAGI" / d).glob("*.persona")) + sorted((home / "cryptoAGI" / d).glob("*.json")) |
| edge = [ |
| ("edge: nameless", {"voice": "hi there", "examples": ["a", 3, True, {"x": 1}, None]}), |
| ("edge: whitespace prompt", {"name": "W", "system_prompt": " \n\t\x1c "}), |
| ("edge: control and quotes", {"name": "Q\"uote", "system": "line1\nline2\t\"q\" \\ \x01\x7f é 字 😀
", "samples": ["\b\f\r"]}), |
| ("edge: title and bio", {"title": "Doc", "bio": " padded bio ", "utterances": "one string"}), |
| ("edge: falsy name", {"name": "", "persona": "fallback", "system_prompt": "", "description": "from description"}), |
| ("edge: numbers as name", {"id": 42, "summary": "s"}), |
| ] |
| cases = [] |
| rnd = random.Random(20260929) |
| for f in files: |
| try: |
| raw = json.loads(f.read_text(encoding="utf-8")) |
| except Exception: |
| continue |
| if not isinstance(raw, dict): |
| continue |
| edge.append((f"{f.parent.name}/{f.name}", raw)) |
| for label, raw in edge: |
| ex = [{"user": e["user"], "assistant": e["assistant"]} for e in (raw.get("exchanges") or []) if isinstance(e, dict) and isinstance(e.get("user"), str) and isinstance(e.get("assistant"), str)] |
| if not ex: |
| ex = [{"user": "Who are you?", "assistant": "Myself."}, {"user": "What now?\n", "assistant": "Ship it, \"carefully\"."}] |
| for seed in (True, False): |
| p = persona_from_dict(raw) |
| rows = build_script_rows(p, [Exchange(**e) for e in ex], seed_voice=seed) |
| with tempfile.TemporaryDirectory() as t: |
| text = write_script_jsonl(rows, Path(t) / "s.jsonl").read_text(encoding="utf-8") |
| cases.append({"label": label, "persona_raw": raw, "persona": p.model_dump(), "exchanges": ex, "seed_voice": seed, |
| "jsonl": text, "params": derive_training_params(len(rows))}) |
| (out / "script.jsonl").write_text("".join(json.dumps(c, ensure_ascii=False) + "\n" for c in cases), encoding="utf-8") |
|
|
| |
| words = ["I", "am", "Savante", "the", "office", "grades", "don't", "won't", "code", "is", "law", "keys", "ARE", "identity", "42", "2026", |
| "İstanbul", "Kelvin", "café", "naïve", "hello", "world", "sovereignty", "verification", "beats", "permission", "—", "!!", "x'y", "a1b2"] |
|
|
|
|
| def utter(): |
| if rnd.random() < 0.05: |
| return "" |
| return " ".join(rnd.choice(words) for _ in range(rnd.randint(1, 14))) + rnd.choice(["", ".", "?", " …"]) |
|
|
|
|
| imp = [] |
| for i in range(3000): |
| n = rnd.randint(0, 6) |
| before = [utter() for _ in range(n)] |
| after = [utter() for _ in range(n if rnd.random() < 0.9 else rnd.randint(0, 6))] |
| baseline = [utter() for _ in range(rnd.randint(0, 5))] |
| r = imprint.score_imprint(imprint.default_inquiries("Savante")[:n], before, after, baseline) |
| imp.append({"before": before, "after": after, "baseline": baseline, "report": r.model_dump(exclude={"inquiries", "before", "after"})}) |
| (out / "imprint.jsonl").write_text("".join(json.dumps(c, ensure_ascii=False) + "\n" for c in imp), encoding="utf-8") |
| print(f"{len(cases)} scripts ({len(edge)} personas and edge cases × seed_voice) and {len(imp)} imprint reports from mindXtrain {MXT} → {out}") |
|
|
